Cairo only needs to support one version of the shader language API,
thanks to the dispatch table. This seems unlikely to change any time
soon. This makes the addition of new features, such as a uniform
location cache, simpler.
In OpenGL ES 2.0, repeat wrap modes (GL_REPEAT and GL_MIRRORED REPEAT) are
only available for NPOT textures if the GL_OES_texture_npot is supported.
This commit adds a shader implementation of these wrap modes for use by
devices that do not support GL_OES_texture_npot.
Having spent the last dev cycle looking at how we could specialize the
compositors for various backends, we once again look for the
commonalities in order to reduce the duplication. In part this is
motivated by the idea that spans is a good interface for both the
existent GL backend and pixman, and so they deserve a dedicated
compositor. xcb/xlib target an identical rendering system and so they
should be using the same compositor, and it should be possible to run
that same compositor locally against pixman to generate reference tests.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
P.S. This brings massive upheaval (read breakage) I've tried delaying in
order to fix as many things as possible but now this one patch does far,
far, far too much. Apologies in advance for breaking your favourite
backend, but trust me in that the end result will be much better. :)
According to the GLSL ES standard, the precision of variables in the fragment
shader should be explicitly defined (although mesa's compiler is not that
strict).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
1D textures are not available in GLES2. We use 2D textures with height 1
instead, which are equivalent to 1D textures and exist across all GL flavors.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The GL_CLAMP_TO_BORDER wrapping method (used with CAIRO_EXTEND_NONE) is not
available in GLES2. We use shaders to implement similar functionality for
GLES2.
If bilinear filtering is used, the shader performs a linear fade to
transparency effect in the texel coordinate intervals [-1/2n, 1/2n]
and [1 - 1/2n, 1 + 1/2n] (n: texture size).
If nearest filtering is used, the shader ensures that a clear color is
used for all texel coordinate values outside [0, 1).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This makes the shader emitter functions more versatile. It allows them to use
all information provided by the operands and the context to decide what shader
to produce.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We never bound any other value to the sampler uniforms, so why bother
resetting them all the time?
[ 0] before firefox-talos-gfx 29.972 30.493 0.89% 3/3
[ 0] after firefox-talos-gfx 29.397 29.599 1.27% 3/3
Signed-off-by: Eric Anholt <eric@anholt.net>
Built-in vertex attributes like gl_Vertex and gl_Color, have been obsoleted
and removed in recent GL versions and they are not supported at all in GLES2.0.
Custom attributes are supported in all GL versions >= 2.0, in GL 1.x with
ARB shader extensions and in GLES2.0.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The built-in gl_ModelViewProjectionMatrix uniform (and others) has been
deprecated and removed in recent GLSL versions and is not supported
at all in GLSL ES. A custom uniform for the same purpose works across
all versions.
[ickle: base _gl_identity_ortho on the 2D variant of glOrtho i.e. with
fixed near/far values of [-1, 1]]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The ftransform() shader function was deprecated and removed in recent
GLSL versions and is not included at all in GLSL ES.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Due to the fact that we fail if the system doesn't support shaders, we
now always have a valid shader implementation.
Reviewed-by: Eric Anholt <eric@anholt.net>
The non-shaders implementation was never tuned for fixed-function GL
implementation, the maintainers are not interested in supporting it,
and the hardware is rather rare at this point. This lets us focus on
the implementation for modern hardware, which needs plenty of work
still.
Reviewed-by: Eric Anholt <eric@anholt.net>
Cairo assumes that radial gradients are drawn as defined in ISO3200
since b661bc8712.
To implement them in GLSL, three different shaders are required:
- tangent circles (a=0 in the parameter equation)
- none-extended gradient (solutions outside [0,1] must be ignored)
- extended gradient
Moreover, texcoords are used instead of gl_FragCoord, to fix the
flipping as per 31e116f084.
Fixes huge-radial, radial-gradient, radial-gradient-extend,
radial-gradient-mask, radial-gradient-source.
The issue is that we store our data flipped based on whether we're
rendering to an FBO or to a window. By not flipping our gl_FragCoord
usage based on that (either with math or ARB_frag_coord_conventions),
this caused linear gradients to be flipped when rendering either to a
window or to an FBO. To avoid this, pass in appropriate texcoords.
And, if we're passing in texcoords, just do the projection to the
linear gradient factor on the CPU side per vertex instead of providing
a bunch of uniforms to do the math per fragment.
Fixes 18 testcases.
In 9b9952ab4f
_cairo_memory_stream_destroy was changed to take an unsigned long
instead of unsigned int, and the two callsites in cairo-gl-shaders.c
weren't updated.
No point in passing in a column of zeroes that we multiply by zero, or
the row that we drop on the floor. We could do even better by just
passing in a 3x2, at the expense of adding a language version
dependency (or just pass in 2 vectors).
No significant performance difference.
And complete the move of the shaders into the cairo_gl_shader namespace.
Of particular note, the bind_*_to_shader become
_cairo_gl_shader_bind_*() and have proper types.
This allows to grow the complexity of the shaders without enlarging the
static array to unreasonable sizes.
The cache size of 64 is essentially random.
The idea is being able to generate shaders from the given input
on-demand. This allows creating more advanced shaders, such as those
that include the destination (for self-painting and
unsupported-by-blending operator) or painting images in weird formats
(extra translation step in shader, think YUV).
I updated the Free Software Foundation address using the following script.
for i in $(git grep Temple | cut -d: -f1 )
do
sed -e 's/59 Temple Place[, -]* Suite 330, Boston, MA *02111-1307[, ]* USA/51 Franklin Street, Suite 500, Boston, MA 02110-1335, USA/' -i "$i"
done
Fixes http://bugs.freedesktop.org/show_bug.cgi?id=21356
Since moving to pixman for the gradient textures, first=0 and last=1,
so there's no need to rescale to 0,1 any more. Shaves 6 Gen4 ISA
instructions, including 2 inverses, in the radial fragment shader.
